Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Welp, there goes a 3nm for the Snapdragon 8 Gen 3 and most likely higher costs passed on to consumers due to Tim Cooks's Apple.

Tim Cook's Apple delivers us the best technology and all you can do is dish up hate and complaints.

Thank you Tim Cook's Apple for working your apple off for us! And TSMC for being an industry leader! I wonder how much Apple's own engineers contributes to TSMC's success rate?

I don't hear much about AMD. Where do they stand with this nanometer downscaling?
 
The industry has already decided that the next node shrink name is going to Ångströms. People seem really confused on this point. There are measurements smaller than nanometer so no we are never going to get to 0 nm. That's silly. The industry is likely to switch over at 1.5 nm and start using 15 Å.
Yes, with some angst the next one will be angstrom
 
  • Like
Reactions: xmach
They're secured the ENTIRE supply? That's nuts. I was about to say, "So, this means that AMD isn't going to get in on the action?" But then I remembered they just launched their 7000 series which I believe is on 5nm, so that makes sense.

Can someone enlighten me on this: I recall seeing someone say something along the lines of TSMC's N3 process not being a "true" 3nm process. Am I taking crazy pills or is there an explanation to that?
 
Apple itself cannot absorb any cost. It is a for-profit corporation. The whole point is to get more money for their product than it cost them producing it.
Lol for profit companies can still absorb costs if they want to. Microsoft lose money on every Xbox and the first one was a net loss.

Apple moving the 3nm might be at a loss at first and they’ll absorb the cost until manufacturing prices come down.
 
  • Like
Reactions: Gudi and mike2q
I just wanted to say props to you, for actually understanding what is going on and what the various technologies and specs mean. I frankly get frustrated reading all the posts with inaccurate info in these forums. I agree with just about everything you have been saying.

I believe the rumor to be reasonable and probably true that apple has purchased all or most of the N3(N3B) production and any new chip released this year will likely be made on that node. N3E will likely not go into high volume production for at least another 6 months if not close to the end of the year if there are any setbacks. So I do not see any new products being released by apple this year based on the N3E process.

But I wanted to add a couple more points:

As you had said, N3E is more relaxed than N3B, this also includes that they have reduced the number of EUV layers from 25 down to 19. This reduced complexity is one of the reasons that yields should also be higher on N3E and as a result the prices(eventually) will be lower despite the slightly lower density.

Also a big advantage that N3E has over N3B is that they introduced FinFlex, this allows for extreme customization/optimization of power and performance by adjusting the number of NMOS and PMOS fins on a cell. With a FinFlex 3-2 implementation it offers 12% lower power iso-performance, 32% higher performance iso-power, and 15% lower area compared to N5(which uses a 2-2 fin design). So compared to N3B an N3E implementation can in part or whole trade some of the power and density improvements to more than double the performance improvements over N5.
Yes, I didn't want to get into the FinFlex stuff, it seems like almost nobody is taking the time to understand the basics, and FF confuses matters even more. FF2-1 is effectively extremely close to N3B, and somewhat cheaper, which is why nobody is going to N3B except Apple - they're willing to wait another 6-9 months to get better pricing for roughly equivalent characteristics, plus the option of making tradeoffs between P, P & A due to FF.

Note that your "prices will be lower" is true so far as it goes. However, if I remember this right, if you're doing FF3-2, your area advantage over N5 is down to <20%, at which point your costs are *higher* per transistor than N5 as the price is about +1/3 per wafer. And that's assuming your yield is as good as on N5.

One thing I don't know - how hard is it to port a design from N3 to N3E using FF2-1? One generation of A processors was made on both TSMC and Samsung; if they were willing to do that, then they might be willing to port A17 from N3 to N3E midstream... maybe. I have no insight into how this would compare.
 
  • Like
Reactions: db604e
I forget who told me, but TSMC outcompeted Intel and AMD, not Apple. Apple is winning by buying up all the supply of TSMC's latest, most advanced chips.
 
  • Like
Reactions: fourthtunz
Exciting, but since I just obtained a MacBook Pro M2Max in full configuration, it will years before I need a replacement. I will wait for the 1nm chips.
Same here. I ain’t looking at Macs no more. These M2 Max systems are zippy. And I mean there are zero hiccups.
 
  • Like
Reactions: MNGR
It's likely to be huge, and even more so for the higher-end chips (Pro/Max/Ultra). The m2 was a stopgap measure due to TSMC's inability to produce N3 on their original schedule. Because of this, Apple couldn't use their original designs for cores on M2 (and A16), instead doing a quick tune-up job on the previous generation cores (from M1 and A15). They had to shelve a lot of progress in their uncore as well, producing very little improvement in scaling. Because of this, the only meaningful advantage of the M2 over the M1 was due to clockspeed and increased core counts.

By contrast, for the M3, they will be able to use the original M2 design, plus all the work they've done since to further progress both in the cores (CPU/GPU/NPU/AMX/etc.) and the uncore. Add to that the possibility of even more cores due to the large scale down of features (somewhat less than you might think, though, due to almost nonexistent SRAM scaling). And finally, they might even bump the clocks more, at least for desktop parts.

In all, it's possible you'll see single-core boosts in the 20% range, but more likely it'll be higher - perhaps as high as 40%+, which is a ridiculously large bump for a single generation. Multicore is likely to see even higher improvements due to more cores, but also due to better efficiency in the uncore, where at least on the larger chips, there's a ton of room for improvement. It's plausible that Apple will again stomp all x86 processors, both mobile and desktop, and not by a little. They should at least hit parity.
I can’t wait for the M3 macs so I can pick up a used M2 Mac! Good times we live in!
 
Welp, there goes a 3nm for the Snapdragon 8 Gen 3 and most likely higher costs passed on to consumers due to Tim Cooks's Apple.

So you’re complaining that they are in the catbird seat vs their biggest competitor due to extremely good planning and also that they turn a profit for the goods and services they bring to market? Did you take classes at the Packard Bell School of Business at Zenith Data Systems University?
 
Apple: "The new chip is more efficient? Okay, let's reduce the battery capacity. Guys, you got all day battery life on your 2k iPhone 15 pro" well done Apple!

Jony Ive would have done that. Not the new guard. They would never give up the chance to say “the phone now lasts two days” or whatever it ends up being.
 
Apple to announce (and possibly release) the ASi Mac Pro with M3 Ultra & M3 Extreme SoCs (using the N3B process) at WWDC 2023...
 
Apple to announce (and possibly release) the ASi Mac Pro with M3 Ultra & M3 Extreme SoCs (using the N3B process) at WWDC 2023...
I hope they don't waste the first batch of 3nm chips on something like VR goggles... Not something I see myself using anytime soon.
 
We know Apple is taking all of TSMC's supply of N3, and it's likely (but NOT certain) that it won't go into the next iphone
Most recent reports state that TSMC's N3 wafer production is 45,000 per month till end of 2023. With Mac sales at approximately 25 million units per year, I highly doubt Mac sales till the end of year can absorb that much wafer capacity. I actually wouldn't be surprised if this year's entire N3 production will go to the iPhone 15 Pro series and we see the first M3 Macs beginning Spring 2024.
 
Is the m3 supposed to be monumental or incremental update. I just got a m2 air and thinking of m3 eh.
its a pretty significant update, both in terms of performance and battery, m2 series is pretty incremental compare to the m1 since it uses the same 5nm node.
 
Most recent reports state that TSMC's N3 wafer production is 45,000 per month till end of 2023. With Mac sales at approximately 25 million units per year, I highly doubt Mac sales till the end of year can absorb that much wafer capacity. I actually wouldn't be surprised if this year's entire N3 production will go to the iPhone 15 Pro series and we see the first M3 Macs beginning Spring 2024.
How many dies are you estimating per wafer?
 
  • Like
Reactions: theorist9
There are about 2000 people at Apple working on Silicon. Even if a hundred of them were to leave, and the reported "brain drain" is way smaller than that, that's only 5%. The brain drain might actually have a positive effect as newer recruits will have new ideas.
new recruits also need to be trained up and brought up to speed, i got the short end of the stick and had to train up 2 new recruits beginning of this year, the time and effort taken away from my regular schedule so much so, that my tasks got delayed.
 
Most recent reports state that TSMC's N3 wafer production is 45,000 per month till end of 2023. With Mac sales at approximately 25 million units per year, I highly doubt Mac sales till the end of year can absorb that much wafer capacity. I actually wouldn't be surprised if this year's entire N3 production will go to the iPhone 15 Pro series and we see the first M3 Macs beginning Spring 2024.
What estimates are using for the number of chips/wafer for the various M3 Mac chips and the iPhone 15 chips?
 
Most recent reports state that TSMC's N3 wafer production is 45,000 per month till end of 2023. With Mac sales at approximately 25 million units per year, I highly doubt Mac sales till the end of year can absorb that much wafer capacity. I actually wouldn't be surprised if this year's entire N3 production will go to the iPhone 15 Pro series and we see the first M3 Macs beginning Spring 2024.
If they're really hitting 45k right now, or any time in the next six months, then you're right. All the Macs and all the VR units (and even, all the pure-GPU chips they might be making) wouldn't soak up all that capacity.
How many dies are you estimating per wafer?
It doesn't really matter... if you got 50 dice off a wafer, that would be more than enough. They're definitely getting more than that. :) Even assuming a Mac Pro uses 4-8 dies, total consumption is too small.

There is still one wildcard here... if they make A17s on *both* N3 and N3E, my scenario still has legs. As I mentioned earlier, though, I don't know how reasonable that is. They did dual designs at least once or twice before (for TSMC and Samsung, for at least the A9 I think), but I'm not sure if the situation is at all analogous.
 
My M1 Max will probably still be way more than "adequate" (and by that I mean still super fast) when the M3 comes to the MBP but I look forward to splurging on an M3 Max anyway because I have no self control 😆
Honestly though, if you want the "big increase" this is probably going to be it.
 
If they're really hitting 45k right now, or any time in the next six months, then you're right. All the Macs and all the VR units (and even, all the pure-GPU chips they might be making) wouldn't soak up all that capacity.

It doesn't really matter... if you got 50 dice off a wafer, that would be more than enough. They're definitely getting more than that. :) Even assuming a Mac Pro uses 4-8 dies, total consumption is too small.

There is still one wildcard here... if they make A17s on *both* N3 and N3E, my scenario still has legs. As I mentioned earlier, though, I don't know how reasonable that is. They did dual designs at least once or twice before (for TSMC and Samsung, for at least the A9 I think), but I'm not sure if the situation is at all analogous.
They said production would started late Dec, and would gradually ramp up to 45k/mo in March. So (assuming DigiTimes is right, which it's usually not) let's do a back-of-the-envelope calculation and say 5k Dec., 15k Jan, 30k Feb, 45k thereafter. Further suppose that it takes 1 month between when a chip is produced and it appears in a sold product (no idea, just guessing). At 50 chips/wafer, that's 23M chips available for products sold through the end of 2023. If we assume 250 chips/wafer at 75% yield, that's 85M chips. The number of chips/wafer for the iPhone 15 chip is probably significantly higher.

I don't know the actual sales volume, but expected sales figures for the iPhone 14 Pro/Max's in Q4 2022 was 85M. Let's assume it's roughly 100M for the iPhone 15 Pro/Max's in Q4 2023, since it goes up every year.

Given this, it seems you'd need a much more specific calculation that accounts for the expected iPhone 15 Pro/Max die size, as well as the die sizes for the M3 chips, to determine this volume of wafers would be enough to cover both the iPhone and the Macs.
 
Last edited:
Wouldn’t the most urgent use for this be in the VR headset?!? I mean, isn’t this like, pivotal for minimizing some of the issues inherent in that device? And they wouldn’t need as many for that as they would for their other offerings. I would think that would be destination numero uno for the chip.
 
Last edited:
  • Disagree
  • Like
Reactions: Gudi and xmach
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.